Buyers increasingly ask AI tools who to hire: “best CRM automation agency for ecommerce”, “who builds websites for small businesses near me”. It is natural to type the same question into ChatGPT and see whether your business comes up. That quick check is a reasonable starting point, but a poor measurement.
This article replaces the popular “30-second test” with a version you can repeat, compare over time and act on. It focuses on running the test and interpreting what you see. For the logging worksheet and crawler settings, use our ChatGPT visibility guide.
Why one prompt misleads
Ask the same question three times and you may get three different answers. Here is a fictional example of what that can look like for one question.
- Example provider 1
- MTI Tech linked
- Example provider 2
- Example provider 2
- Example provider 3
- Example provider 1
- Example provider 1
- MTI Tech named, no link
- Example provider 4
Answers vary with the exact wording, the user’s location, whether search is used, the date, which sources are available at that moment and the conversation before the question. A single favourable answer isn’t a win, and a single missing mention isn’t a loss. Patterns across repeated runs are what matter.
The repeatable test
- Fix a question set.
Choose the questions real buyers ask, across category, local, problem, comparison and cost questions. Eight is a workable start; many businesses grow to 10 to 30. Keep the wording identical every time.
- Record the context.
Note the platform and mode, the date, your location settings and whether you were signed in. Start each run in a fresh conversation.
- Log four things per answer.
Were you mentioned? Was your site linked? Which page was cited? Was the description accurate? Keep the cited source URLs.
- Repeat on a schedule.
Monthly is enough for most businesses. Run each question more than once per session if answers vary a lot.
- Compare trends, not screenshots.
Report how often you appear across the set, and alongside it the referral visits from chatgpt.com in your analytics.
Build your question set
Enter your details to generate a starting set of questions across the five types. Edit them to match how your customers actually talk, then keep them fixed.
Reading your results
Choose what you observed to see what it may mean, what it doesn’t prove and what to check next.
An example log
Here is a fictional three-month summary for a business running eight questions monthly. Select a month to see that run. The point is the trend across months, not any single row.
| Question type | Mentioned | Linked | Cited page | Accurate |
|---|
Fictional data for illustration. In this example, the September improvement followed new service pages and a published case study, which is a reason to keep watching, not proof of cause.
What helps you get referenced
You can’t control what an AI system recommends, but you can make accurate information about your business easier to find and use. The same foundations help in ordinary search.
“Website development for small businesses”, not “digital solutions for modern brands”. Name your services, customers and service area.
Scope, process, cost factors, proof and the questions buyers ask, one service per page.
The same business name, contact details and description on your site, profiles and directories.
Case studies with real results, genuine reviews and project examples. Claims on your own site count for less than evidence.
How to choose a provider, what it costs, what to check before hiring: content that helps a decision.
ChatGPT search uses OAI-SearchBot, which is separate from the GPTBot training crawler. Check that your public pages aren’t blocked.
Mistakes to avoid
- Treating one screenshot as proof of success or failure.
- Changing the question wording between runs, then comparing results.
- Assuming absence means lost customers. It is a visibility signal, not a demand measurement.
- Assuming absence means a technical fault. Check access, but most causes are content and evidence.
- Expecting schema markup alone to earn mentions. It helps understanding; it guarantees nothing.
- Trying to game answers with fake reviews or mass mentions. That undermines the trust you need.